Records all three void attempts in Part D rather than quietly dropping them, with the fault each one exposed: production replacing compartments instead of adding (extinction, caught by the parasite column reading zero); a K1 control calibrated identically to its own arm (caught by a T0 guard added after); and payoff made expensive to rediscover with templates still collapsing (T0 again). THE WALL, STATED MORE SHARPLY: templates sit at 0.006-0.010 in every arm of every attempt. A template costs a slot NOW, lowering current function, and between-compartment selection scores current function EVERY generation. Its benefit is deferred and conditional — it pays only if a type is later lost. Making rediscovery expensive raises the deferred payoff without touching the immediate cost. AND THE OBVIOUS REMEDY COLLIDES WITH A MEASURED RESULT: for the deferred benefit to land inside the selection horizon, losses must be frequent, i.e. few copies per type, i.e. small compartments — exactly the regime fidfloor.c measured as destroying sequence information. Fourth time the answer comes back "small compartments correct but forget". Proven/open gains it as a DIAGNOSIS marked with a warning, not a result, since all three runs were void. NEXT STEP RECORDED, and it does not collide with anything measured: make the environment FLUCTUATE. If the target set changes over time, a type not needed now becomes needed later, so holding it in a template pays REGULARLY rather than CONDITIONALLY — converting a deferred benefit into a recurring one without requiring small compartments. Unbuilt. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>